The paper discusses the statistical principles behind iris recognition, emphasizing the role of randomness and the use of wavelets to encode iris patterns. The key to iris recognition is the failure of a statistical independence test, which is virtually guaranteed to pass when comparing phase codes of different eyes but uniquely fails when comparing a person's own phase code with a different version of itself. This test is implemented using Boolean operations on 2048-bit phase vectors, masked by corresponding mask vectors to exclude non-iris artifacts. The resulting fractional Hamming distance (HD) measures the dissimilarity between two irises, with a mean HD of 0.499 for different irises and 0.507 for genetically identical ones. The HD distribution follows a binomial model with 249 degrees of freedom, indicating that the probability of two different irises disagreeing in fewer than a third of their bits is extremely low. This statistical model ensures high accuracy in iris recognition, enabling real-time identification with great precision. The paper also discusses the robustness of iris recognition against variations in size, position, and orientation, and the use of permutation techniques to prevent replay attacks. The algorithm's efficiency is demonstrated through high-speed comparisons, and the paper concludes with the importance of randomness in ensuring the reliability of iris recognition.The paper discusses the statistical principles behind iris recognition, emphasizing the role of randomness and the use of wavelets to encode iris patterns. The key to iris recognition is the failure of a statistical independence test, which is virtually guaranteed to pass when comparing phase codes of different eyes but uniquely fails when comparing a person's own phase code with a different version of itself. This test is implemented using Boolean operations on 2048-bit phase vectors, masked by corresponding mask vectors to exclude non-iris artifacts. The resulting fractional Hamming distance (HD) measures the dissimilarity between two irises, with a mean HD of 0.499 for different irises and 0.507 for genetically identical ones. The HD distribution follows a binomial model with 249 degrees of freedom, indicating that the probability of two different irises disagreeing in fewer than a third of their bits is extremely low. This statistical model ensures high accuracy in iris recognition, enabling real-time identification with great precision. The paper also discusses the robustness of iris recognition against variations in size, position, and orientation, and the use of permutation techniques to prevent replay attacks. The algorithm's efficiency is demonstrated through high-speed comparisons, and the paper concludes with the importance of randomness in ensuring the reliability of iris recognition.